Abstract
We investigate factors affecting women’s decisions to terminate pregnancies in Matlab, Bangladesh, using logistic regression on high-quality data from the Demographic Surveillance System on more than 215,000 pregnancies that occurred between 1978 and 2008. Variables associated with the desire not to have another birth soon (very young and older maternal age, a greater number of living children, the recent birth of twins or of a son, a short interval since a recent live birth) are associated with a greater likelihood of pregnancy termination, and the effects of many of these explanatory variables are stronger in more recent years. Women are less likely to terminate a pregnancy if they don’t have any living sons or recently experienced a miscarriage, a stillbirth, or the death of a child. The higher the woman’s level of education, the more likely she is to terminate a pregnancy. Between 1982 and the mid-2000s, pregnancy termination was significantly less likely in the area of Matlab with better family planning services.
Introduction
Worldwide, about 42 million pregnancies are terminated each year. Many of these terminations are unsafe and result in maternal deaths or disability. Each year, about 70,000 maternal deaths—one in eight maternal deaths worldwide—result from illegal abortions, and such terminations are estimated to lead to 5 million disabilities per year globally, some of which are permanent (Grimes et al. 2006; Shah and Åhman 2009). However, relatively few studies have addressed the influences on decisions to terminate pregnancies, especially in developing countries.
In earlier research (Rahman et al. 2001), we analyzed the effects of family planning (FP) services on rates of pregnancy termination in Matlab, a typical rural subdistrict of Bangladesh, for the years 1979–1998. Here, we extend our research to consider a longer and more recent period of time (1978–2008) and to analyze the effects of socioeconomic, demographic, and programmatic variables in a multivariate analysis of decisions to terminate pregnancies.
Conceptual Framework
The likelihood of a live birth (LB) and hence the fertility of women (e.g., children ever-born) and of populations (e.g., fertility rates) are the result of three processes: sexual intercourse, conception, and a pregnancy that results in a LB (Bongaarts 1978). Each process is influenced by behavioral, physiological, and exogenous factors. For example, given that coitus occurs, the likelihood of conception is affected by the use of contraception and the woman’s fecundity. Given that conception occurs, the likelihood that an embryo will result in a LB is affected by a decision to intentionally terminate the pregnancy (induced abortion); physiological factors may affect the likelihood of a miscarriage or stillbirth.
Couples typically have a target number of living children they would like to have, and they may care about the timing and spacing of births and the sex composition of their children. The number of children desired is affected by the costs and benefits that couples associate with children. If parents experience a child’s death, they may have an additional birth to replace the child who died (Preston 1978).
Couples use various behaviors in an effort to achieve their desired number and timing of pregnancies. However, some of these behaviors, particularly sexual intercourse, are desired for other reasons as well, and may result in unintended pregnancies if effective contraception is not used when pregnancy is not desired. Because contraception has costs (information, monetary, psychic, and possible physical discomfort), women who do not want to become pregnant do not necessarily practice (effective) contraception. After conception occurs, women can decide whether to terminate the pregnancy or see it to term. Pregnancy termination is unlikely if the pregnancy is intended, but is possible if the woman does not want to have more children or does not want a child at that time, and is more likely when the “costs” of an unintended pregnancy are higher. Like contraception, pregnancy termination has information, monetary, and psychic costs, as well as the possibility of serious health consequences for women (Bhuiya et al. 2001).
In this article, we look at influences on the likelihood that pregnancies are voluntarily terminated. We use a sample of pregnancies, so we are dealing with cases that have already experienced the first two processes of producing a LB: intercourse and conception. Women in couples who do not want (more) children and do not have sexual intercourse or who successfully prevent conception through the use of contraceptives are not in our sample. Our sample consists of a combination of intended and unintended pregnancies. In explaining whether pregnancies are voluntarily terminated, we are explaining the net result of the incidence of unintended pregnancy and the likelihood that those unintended pregnancies are voluntarily terminated. Our presumption is that if a pregnancy is not intentionally terminated, the woman has chosen to take the pregnancy to term and have a LB. We recognize that a small proportion of pregnancies result in miscarriages or stillbirths. However, voluntary pregnancy termination (VPT)1 occurs, on average, before all other types of pregnancy outcomes,2 and hence decisions about VPT typically occur before the competing risks of these other outcomes.
We hypothesize that variables associated with lower fertility and the desire not to have another birth (soon) will be correlated with a greater likelihood of VPT; for example, older women or those with many living children may have completed their family size and may opt for VPT if they become pregnant. Later in the article, we discuss the specific variables we consider.
Pregnancy Termination in Bangladesh
Bangladesh is a poor, traditional, religiously conservative nation, with a small geographic area but a large population (152.9 million in 2012) and one of the highest population densities in the world (1,062 persons per km2) (Population Reference Bureau 2012). Since independence in 1971, Bangladesh has had a strong political commitment to reduce its high rate of population growth. In the past four decades, fertility in Bangladesh has fallen markedly, from a total fertility rate (TFR) of 6.3 children per woman in the early 1970s to 3.3 in the late 1990s and 2.3 in 2011 (National Institute of Population Research and Training (NIPORT), Mitra and Associates, and MEASURE DHS ICF International 2013).
Early-gestation pregnancy termination is legal in Bangladesh if performed in a medical setting before pregnancy is clinically confirmed. Such terminations are done by manual vacuum aspiration by trained female paramedics at government Health and Family Welfare Centers and are known as “menstrual regulation” (MR). MR can be performed only within 10 weeks of the last menstrual period. The consent of the woman’s husband is not officially required, but many providers ask for it. MR has been available through government and other medical facilities in Bangladesh since the late 1970s, when the government agreed to permit such pregnancy terminations in an effort to replace the practice of unsafe pregnancy termination.3
Pregnancy termination in a nonmedical setting or after pregnancy is clinically confirmed is prohibited in Bangladesh except when done to save a woman’s life. Nevertheless, unsafe terminations, available from traditional healers (usually older women who perform the procedure by inserting herbal roots or other solid objects into the uterus), appear to be common and have been found to be a leading cause of maternal mortality and short- and long-term maternal health complications (Dixon-Mueller 1988; Huda et al. 2010; Ronsmans et al. 1997).
Study Area
We use data from Matlab, a subdistrict typical of rural Bangladesh that is well known for its Demographic Surveillance System (DSS), operated by the International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b) since 1966. Since 1977, Matlab has been the site of an FP initiative, in which the Maternal Child Health and Family Planning Project has provided one-half of the Matlab study area, referred to as the MCH-FP Area, with more accessible and higher-quality FP services than the standard government services available in the other half, known as the Comparison Area. The Comparison Area is typical of much of Bangladesh in contraceptive practice (icddr,b 2011), fertility (Mitra et al. 1994; NIPORT, Mitra and Associates, and ORC Macro 2009), abortion (Khan et al. 1986b), and maternal mortality (Alauddin 1986; Khan et al. 1986a; NIPORT, ORC Macro, Johns Hopkins University, and ICDDR,B 2003). The mean desired number of children has been similar in both areas and has decreased at similar rates, from about 4.5 in 1975 to 3.0 in 1990 to 2.5 in 2000 (Bairagi 2001; Koenig et al. 1992). The two areas had the same TFRs—2.5 births per woman—in 2009 (icddr,b 2011).
At least until recently, the MCH-FP Area has been characterized by greater contact among clients, workers, and supervisors, as well as by greater availability and a broader mix of contraceptive methods than available in the Comparison Area (Koenig et al. 1992). From 1977 to 1998, specially trained female community health workers (CHWs) in the MCH-FP Area visited married women of reproductive age (MWRA) every two weeks to provide counseling about FP services and to deliver oral contraceptives, condoms, and injectable contraceptives at the doorstep. In the Comparison Area, MWRA were supposed to receive doorstep visits every two months from female welfare assistants of the government FP program, although evidence suggests these visits didn’t always occur. In 1999, visits by CHWs to women in the MCH-FP Area were reduced to once monthly. In 2000, FP services in the Comparison Area were delivered from fixed-site clinics rather than doorstep visits. In 2001, CHW visits to women in the MCH-FP Area were eliminated, and FP services in this area also were provided from fixed-site facilities. In addition to the standard government Health and Family Welfare Centres, the MCH-FP Area also has icddr,b subcenters that provide MCH and FP services. Services provided by icddr,b facilities are of better quality than those provided by the standard government facilities (Koenig et al. 1997).
The difference in contraceptive services between the two areas has led to a difference in contraceptive practice; for example, in 2008, 54 % of MWRA in the MCH-FP Area used contraception, compared with 41 % in the Comparison Area.4 These areas also differ in the contraceptives used (icddr,b 2011). Users in the MCH-FP Area are much more likely to use injectables, whereas those in the Comparison Area are more likely to use oral contraceptives (icddr,b 2011).
At its beginning in 1977, the MCH-FP project provided MRs as a backup strategy in case of contraceptive failure, in addition to those offered by government clinics in both areas (Bhatia and Ruzicka 1980). This was discontinued in 1983, when donors withdrew their support from that part of the program.
Data and Methods
Data from the Matlab DSS
Until 1997, specially trained female CHWs employed by the DSS visited every household in both areas every two weeks to record any pregnancy outcomes occurring since the previous visit. These visits occurred monthly beginning in 1998 and have been bimonthly since 2007. A study comparing the reporting of vital events for fortnightly, monthly, and bimonthly recall periods concluded that the accuracy of reporting was the same in all three systems (Alam et al. 1999).
The DSS has distinguished induced abortions from spontaneous abortions (miscarriages) since 1977. Our multivariate analysis of influences on induced abortion considers the period April 1978–December 2008. (We do not consider January 1977–March 1978 in the regression analysis because, as described later in this article, we allow the effects of durations of interpregnancy intervals shorter than 15 months to differ by the type of outcome with which they began. We can define these interactions for intervals that began in January 1977, but intervals 15 months long didn’t end until March 1978.)
The Matlab data on pregnancy outcomes are likely to be of high quality and are not likely to suffer from underreporting by women declining to report VPT for personal, familial, social, or religious reasons. In their many years of work in the community, the CHWs have established themselves as trustworthy and are thus in a good position to collect reliable information on pregnancy and VPT. In addition, because of the frequency of their visits, they were likely to know pregnancy status and changes. However, a recent investigation we conducted using the data on maternal deaths showed that in some cases the death records listed the cause of death as abortion, but the DSS records on the pregnancies did not show them as cases of abortion (instead coding them as “deaths during pregnancy”), suggesting that there may be some underreporting of VPTs in the DSS.
Sample and Statistical Analyses
We use Matlab DSS data on 215,256 pregnancies that occurred between April 1978 and December 2008. Of these, 7,594 (3.53 %) were voluntarily terminated.
Descriptive Analysis
We begin with a descriptive analysis that shows the 1979–2008 trend in each area in the general abortion rate (GAR), which is the number of induced abortions per 1,000 women of reproductive age per year. For this study, we consider reproductive age to be 10–54 because we observe pregnancies and VPTs occurring for women younger and older than the typically used ages of 15–49. (We have also computed the GAR for women aged 15–49 and see the same patterns as for ages 10–54.)
A relatively high GAR for a particular area or year can reflect that many women in that group became pregnant or that women who became pregnant were likely to terminate their pregnancies. To distinguish between these possibilities, we also examine pregnancy rates (pregnancies per 1,000 women) and the proportions of pregnancies that are terminated (VPTs per 1,000 pregnancies). Multiplied together, the pregnancy rate and the proportion of pregnancies terminated produce the GAR.
For each year, we use t tests to assess whether differences between areas are statistically significant.
Multivariate Analysis
We estimate a logistic regression to assess the effects of each of the explanatory variables that we consider on whether the pregnancy was terminated or instead resulted in an LB, stillbirth, or miscarriage.5 As noted earlier, on average, VPT occurs before all other types of pregnancy outcomes. Our estimation uses the vce option in STATA to adjust standard errors for the fact that we have more than one observation on some women. (The 215,256 pregnancies in our sample occurred to 76,648 women.)
Explanatory Variables and the Hypotheses They Test
Following our conceptual framework, our explanatory variables are measures of factors that affect the likelihood that a woman experienced an unintended pregnancy and chose to terminate that pregnancy. The explanatory variables we consider and the hypotheses they test are described in the following sections. We explore the possibility that the effects of the explanatory variables vary over time by allowing the effects of all explanatory variables to differ between two time periods: 1978–2000 and 2001–2008.6Appendix B shows the means for these two periods of the explanatory variables we consider and tests whether they differ significantly between the periods.
The Woman’s Age
We consider age groups 15 or younger, 16–17, 18–19, 20–24, 25–29 (reference category), 30–34, 35–39, 40–44, and 45 and older. We expect that among pregnant women, termination probabilities will be highest for the youngest and oldest women because they are least likely to want to have a(nother) child. In both time periods, the largest percentage of pregnancies occurs at ages 20–24, followed by ages 25–29. The rank ordering of age groups is the same in both periods, although there are significant differences in the proportions at most ages because of our large sample.
Number of Living Children at Time of Index Pregnancy
We consider the following categories of number of living children at the time of the index pregnancy: 0 (reference category), 1, 2, 3, 4–5, and 6 or more. We expect that, all other things the same, the more children the woman already has, the less likely she is to want more, and hence the more likely she should be to terminate a pregnancy should one occur. In the 1978–2000 period, 27.6 % of the pregnancies in our sample were to women with no living children, and 39.3 % were to women with one to two living children; by 2001–2008, these percentages had increased to 35.2 % and 47.0 %, respectively. Correspondingly, the percentage of pregnancies to women with four or more living children fell considerably between the periods, from 20.1 % to 7.5 %.
No Living Sons
We include a dichotomous indicator for whether the couple has no living sons. Because couples in Bangladesh have a preference for sons (e.g., Rahman et al. 1992), ceteris paribus, we expect them to be less likely to terminate a pregnancy if they have no living sons. In the 1978–2000 period, 46.1 % of pregnancies in our sample were to women without a living son; by 2001–2008, this had increased significantly, to 56.5 %.
Duration of the Preceding Interpregnancy Interval, and Its Interaction With the Pregnancy Outcome at the Beginning of the Interval and With the Sex of the Child Born Then
Parents have preferences about the timing and spacing of births and generally do not want to have LBs spaced too closely together (e.g., NIPORT, Mitra and Associates, and ORC Macro 2009, for Bangladesh). Hence, we hypothesize that women may be more likely to terminate pregnancies that come sooner than desired after the preceding LB. Because of gender preference, this should be more likely if the previous pregnancy resulted in the birth of a son. Very long intervals may indicate that the woman did not want to become pregnant again, and hence pregnancies after long intervals may also be more likely to be terminated.
If the previous pregnancy did not result in a LB when one presumably was wanted, the couple should be less likely to terminate the index pregnancy. Hence, a pregnancy following a preceding pregnancy that resulted in a miscarriage or stillbirth should be less likely to be terminated than a pregnancy following one that ended in a LB.
If the previous pregnancy ended in VPT, we expect a high likelihood that the index pregnancy will be terminated also because (1) the woman did not want to have a child the last time she was pregnant and so probably doesn’t want a child now, either; and (2) she has already demonstrated that she is willing to have a VPT.
We do not consider the outcome of the previous pregnancy for interpregnancy intervals (IPIs) longer than 14 months because relatively few intervals longer than 14 months began with a non-LB (DaVanzo et al. 2007).
We consider the following IPI-related variables: 7
IPI < 6 months, previous outcome was LB of a male
IPI < 6 months, previous outcome was LB of a female
IPI < 6 months, previous outcome was stillbirth
IPI < 6 months, previous outcome was miscarriage
IPI < 6 months, previous outcome was VPT
The same as the five above, but for IPIs of 6–14 months
IPI = 15–26 months
IPI = 27–50 months (reference category)
IPI = 51–74 months
IPI = 75 or more months
Duration of IPI unknown
First Pregnancy
IPIs were longer in the 2001–2008 period than in 1978–2000. For example, IPIs of 51 months or more made up 18.8 % of IPIs in the later period but only 8.5 % in the earlier period.
Previous Outcome Was Multiple Birth
We include a dichotomous indicator of whether the previous pregnancy resulted in twins or triplets. We expect that women who recently gave birth to more than one child are less likely to want another child. Multiple births are rare in our sample (0.7 % of outcomes in 1978–2000, 0.5 % in 2001–2008).
Previous Child Death and Sex of Child Who Died
We consider a dichotomous variable that indicates whether any of the woman’s children died since her most recent previous pregnancy outcome. We hypothesize that women who become pregnant after a child’s death are trying to replace the child who died, and hence the pregnancy is more likely to be intended and less likely to be terminated. We distinguish the sex of the child who died. Because of son preference, we expect women to be especially unlikely to terminate a subsequent pregnancy if the child who died was a boy. The percentage of women who experienced a recent child’s death fell markedly between the two periods we consider, from 8.4 % in 1978–2000 to 3.7 % in 2000–2008, reflecting the decline in infant and child mortality over the study period.
The Woman’s Education
In Bangladesh, more-educated women are less likely than those with less education to use permanent methods of contraception (tubectomy and vasectomy for their husbands) and are more likely to use nonpermanent methods (condoms, oral contraceptives, or traditional methods (mostly periodic abstinence and withdrawal)) that have higher failure rates (Khan and Rahman 1997). Hence, educated women may be more likely to have an unintended pregnancy. Furthermore, educated women may perceive higher “costs” of having an unintended birth; for example, they are more likely to see the importance (and costs) of educating a child. In addition, educated women are likely to have greater decision-making power and better access to information (e.g., regarding MR) (NIPORT, Mitra and Associates, and ORC Macro 2009). For all these reasons, we expect the likelihood of termination to be positively related to the woman’s level of education.
We consider four categories of the woman’s education—0 (reference category), 1–5, 6–10, and 11–16 years of schooling—and a dichotomous indicator for level of education unknown. In the 1978–2000 period, 57.0 % of the pregnancies in our sample were to women with no education; by 2001–2008, this number had fallen to 26.1 %. The percentage of pregnancies to women with 6–10 years of schooling increased dramatically over this period, from 11.3 % to 41.2 %. The changes in women’s education between the time periods are the largest of any of the explanatory variables we consider.
Husband’s Education
Husband’s education should have effects similar to the wife’s. We consider the same educational categories as we do for women. There was a decrease between 1978–2000 and 2001–2008 in the percentage of pregnancies in our sample that are to women whose husbands have no education, though this is the modal group in both time periods. The changes over time in husbands’ education are smaller than in wives’ education.
Household Space
We expect the likelihood of termination to be positively associated with economic status because wealthy couples are more likely to use less-effective methods of contraception (Khan and Rahman 1997), and they are likely to have more information and can better afford the costs of a VPT.
We categorize the household space of all rooms, a commonly used indicator of household well-being in studies of Matlab (collected in periodic DSS censuses), into four groups approximating quartiles of the distribution, measured in square feet: < 170 (low), 170–249 (low-medium), 250–349 (medium), and 350 or more (medium-high); we also include a dichotomous indicator for household space unknown. Relatively more pregnancies occurred to women with the lowest level of household space in the later time period than in the earlier period.
Religion
Nearly 90 % of the people in Matlab are Muslim (icddr,b 2011); most of the rest are Hindu. The latter are comparatively less conservative and attach less of a stigma to abortion (Ahmed et al. 1996), so we expect them to have a higher likelihood of termination than Muslims. The percentage of pregnancies to Muslim women increased slightly (but significantly) between the two time periods, from 87.6 % to 89.6 %, reflecting the slower decline in Muslim fertility (Rahman and Sutradhar 1996).
Area of Matlab
Rates of contraceptive use are higher, and relatively more use is of effective methods, in the MCH-FP Area than in the Comparison Area. Hence, more of the pregnancies in the MCH-FP Area were intended (Rahman et al. 2001). In recent years, fertility has reached a comparable level in the two areas because of the larger increase in contraceptive use in the Comparison Area (icddr,b 2011). Hence, it is possible that the difference in incidence of VPT between the areas has become smaller or nonexistent. In the 1978–2000 period, there were more pregnancies in the Comparison Area than in the MCH-FP Area, reflecting the higher fertility rates in the former. By 2000–2008, the percentages were nearly the same in the two areas, reflecting the greater fertility decline in the Comparison Area.
Calendar Year
We include dichotomous indicators for five subperiods: 1978–1982 (reference category), 1983–1992, 1993–2000, 2001–2006, and 2007–2008.8 The MCH-FP project provided MR as a backup solution in case of contraceptive failure during the 1978–1982 subperiod but not thereafter. The 1983–1992 subperiod is when the “Mexico City Policy” was first in effect. This policy required nongovernmental organizations to agree, as a condition of their receipt of U.S. government funds for international FP programs, that they would neither perform nor actively promote abortion, even with their own funds. This policy was not in effect during the 1993–2000 subperiod. It was reinstated in 2001. The Mexico City Policy is intended to reduce the incidence of VPT, but it might actually increase it if it leads to reductions in contraceptive use and these reductions, in turn, lead to more unintended pregnancies (Bendavid et al. 2011).
In addition, we expect that the likelihood of VPT has increased over time in Bangladesh as children became more expensive (e.g., parents increasingly recognized the importance of educating their children), and hence the costs of having an unintended child increased. The increasing availability of MR, a legal and relatively safe method of VPT, reduced the total costs of VPT, which may increase the likelihood that pregnancies were terminated.
Interactions Between Area and Calendar Year
We include interactions of area and calendar year to allow the area differences to vary over time. As noted, MR was provided in the MCH-FP Area prior to 1982, so we expect the difference between areas to be different for that subperiod. Furthermore, because the differences in contraceptive use between the areas have shrunk over time; the difference in VPT incidence between them is likely to have diminished as well.
Results
Descriptive Analysis
General abortion rates (GARs) were higher in the MCH-FP Area than in the Comparison Area between 1979 and 1981 (Fig. 1); these were years during which the MCH-FP project provided MR as a backup option in case of contraceptive failure. In every year between 1982 and 2007, the GAR was larger in the Comparison Area than in the MCH-FP Area; the differences are large and statistically significant for the years 1983–2007. Between 1984 and 2004, the GAR was generally two to three times larger in the Comparison Area than in the MCH-FP Area, with the ratio reaching a high of 3.7 to 1 in 1997. The GAR increased fairly consistently between 1983 and 2003 in the Comparison Area, whereas it did not vary much over this period in the MCH-FP Area. However, the GAR rose between 2005 and 2008 in the MCH-FP Area, whereas it fell after 2003 in the Comparison Area. In 2008, the GAR in the MCH-FP Area exceeded that in the Comparison Area, but the difference is not statistically significant.
In Figs. 2 and 3, we investigate whether the patterns in GARs in Fig. 1 are due to differences by time or area in pregnancy rates or in the probability that pregnancies are terminated.
As shown in Fig. 2, pregnancy rates have generally declined over time in both areas. Between 1979 and 2000, the pregnancy rate was about 20 % to 40 % higher in the Comparison Area than in the MCH-FP Area. The relative difference was largest, at 44 %, in 1990 and generally decreased thereafter. The difference between the areas was statistically significant in each year between 1979 and 2005. By 2008, however, the pregnancy rate was significantly lower in the Comparison Area than in the MCH-FP Area (p < .01); this explains the GAR difference in that year.
The likelihood pregnancies are terminated (VPTs per 1,000 pregnancies), shown in Fig. 3, displays a pattern similar to that seen for the GAR in Fig. 1. The likelihood of termination in the MCH-FP Area exceeds that in the Comparison Area for the years 1979–1983, and by an even greater extent than the GAR; VPTs per 1,000 pregnancies are significantly higher in the MCH-FP Area in 1979, 1980, and 1981. The difference is greater for VPTs per 1,000 pregnancies than for the GAR because the pregnancy rate was lower in the MCH-FP Area than in the Comparison Area during those years. Between 1984 and 2007, the relative differences between the areas are greater for the GAR than for VPTs per 1,000 pregnancies because areal differences in pregnancy rates reinforce those in VPTs per 1,000 pregnancies in those years. Nonetheless, between 1984 and 2006, pregnancies were 50 % to 220 % more likely to be terminated in the Comparison Area than in the MCH-FP Area, and the differences are statistically significant at p < .05 for each year between 1984 and 2007. The likelihood of VPT is essentially the same in both areas in 2008.
We now discuss our multivariate analysis of whether pregnancies are terminated (Table 1).
Logistic Regression Analysis
Table 1 shows odds ratios (ORs) for 1978–2000 and 2001–2008 and indicates their level of statistical significance. The final column shows the statistical significance of the differences between the 1978–2000 and 2001–2008 odds ratios.
The Woman’s Age
Consistent with our hypothesis (and as seen in other studies; e.g., Ahmed et al. 1998), woman’s age has a U-shaped relationship with the likelihood of VPT; this is the case in the 1978–2000 and 2001–2008 periods. Although there are relatively few pregnancies to women aged 15 or younger or 45 and older, if pregnancies occur at those ages, the odds of VPT are particularly high. In both periods, the odds of VPT are highest for women age 15 or younger (OR = 5.29 (p < .0001) in the earlier period; OR = 11.11 (p < .0001) in the later period) compared with those aged 25–29 (reference group). The odds are also high for those aged 45 and older (OR = 3.88 (p < .0001) in the earlier period; OR = 6.12 (p < .0001) in the later period).9 Except for ages 30–34 in 2001–2008, all age groups considered have significantly higher likelihoods of VPT than the reference group. The relationship with maternal age is considerably stronger in the 2001–2008 period than in the 1978–2000 period (and most of the interperiod differences are statistically significant), suggesting that in more recent years, women associated greater costs with having an unwanted child at what they view as a too-young or too-old age. Ages 20–29 appear to be the preferred ages for childbearing in the earlier time period, whereas ages 25–34 appear to be the preferred ages in 2001–2008.
Number of Living Children at the Time of the Index Pregnancy
In the 1978–2000 period, beyond one child, the more living children the woman already has, the more likely she is to terminate a pregnancy should one occur.10 Other studies have also found a general increase in the likelihood of pregnancy termination with the number of living children (e.g., Ahmed et al. 1998). In the 2001–2008 period, the odds ratio increases monotonically with the number of living children, and the relationship is much stronger than in the earlier time period. For example, in 1978–2000, the odds of VPT for women with six or more living children is 2.72 times that for women with no living children; in 2001–2008, the corresponding odds ratio is 17.7.
No Living Sons
As expected, in both time periods, women are significantly less likely to terminate a pregnancy if they have no living sons than if they have some. This relationship is somewhat stronger in the later time period (p < .05).
Duration of the Preceding IPI and Its Interaction With the Outcome of the Pregnancy at the Beginning of the Interval and With the Sex of the Child Born Then
As expected, we find that pregnancies after short IPIs that began with a miscarriage or stillbirth are less likely to be terminated than those following IPIs that began with a LB, whereas those following IPIs that began with a VPT have a very high likelihood of being terminated (in the 1978–2000 period, OR = 3.64 for IPIs of less than 6 months that began with a VPT; OR = 3.99 for IPIs of 6–14 months that began with a VPT, each compared with IPIs = 27–50 months). DaVanzo et al. (2007) reported similar findings. The patterns are similar but not as strong in the 2001–2008 period.
Consistent with our hypothesis about son preference, pregnancies following short IPIs that began with the LB of a son are significantly more likely to be terminated than those that began with the LB of a daughter (p < .001 for the sex differences for IPIs less than 6 months and IPIs of 6–14 months); this is true in both time periods.
Pregnancies following IPIs that began with a LB and are less than 6 months long are more likely to be terminated than those following IPIs of 6–14 months, which in the 1978–2000 period are more likely to be terminated than those following IPIs of 15–26 months. These findings suggest that parents do not want to have births spaced closely together, especially after the birth of a son. In the 1978–2000 period, there is no difference in the effects of IPIs of 15–26 and 27–50 months, but the risk increases for intervals longer than 50 months. In the 2001–2008 period, the likelihood of VPT is lowest for IPIs of 27–50 and 51–74 months, and these effects do not differ significantly from each other. In both periods, the likelihood of VPT is elevated for the longest intervals, especially in the earlier period.
Previous Outcome Was Multiple Birth
In the 2001–2008 period, if the previous pregnancy resulted in multiple births, the odds ratio of a VPT is 72 % higher than if the previous outcome was a singleton LB. The effect of multiple births is not statistically significant in the earlier time period.
Previous Child Death and Sex of Child Who Died
If the woman experienced the death of any of her children since her most recent previous pregnancy outcome, she is much less likely to terminate a pregnancy than if she didn’t lose a child (ORs = 0.18–0.33). The likelihood is lower if the child who died was a son rather than a daughter; this sex difference, which is statistically significant (p < .001) in both time periods, is larger in magnitude in the earlier time period.
Woman’s Education
As expected, the higher the woman’s level of education, the more likely she is to terminate a pregnancy. The relationship is stronger in the earlier time period than in the later period. In the 1978–2000 period, the odds for women with the highest level of education (11–16 years) is 2.46 times that for women with no education; the comparable figure for 2001–2008 is 1.93.
Husband’s Education
The more educated the husband, the more likely his wife is to have a VPT. The effect of the husband’s education is not as strong as the effect of the wife’s. These patterns are seen in both time periods, with no significant differences between the two.
Household Space
In the earlier time period, women in the highest category are significantly more likely to terminate their pregnancies. There is no difference among the household-space categories in the later time period.
Religion
Like Ahmed et al. (2005), we find that non-Muslims are significantly more likely to terminate a pregnancy than are Muslims. However, the difference is statistically significant only in the earlier time period (OR = 1.21; p < .0001).
MCH-FP Area and Calendar Year
During the 1978–1982 subperiod, pregnancies were much more likely to be terminated in the MCH-FP Area than in the Comparison Area (OR = 1.74, p < .0001). In all subsequent subperiods through 2006, pregnancy termination is significantly less likely in the MCH-FP Area (p < .0001 in each subperiod). There is no difference between the areas in 2007–2008.
The trends over time in Table 1 are confounded by the fact that we allowed the coefficients of all explanatory variables to differ between 1978–2000 and 2001–2008. Table 2 shows odds ratios for the Year variables from a specification that does not allow the effects of the explanatory variables to differ between the two time periods. In the Comparison Area, the likelihood of VPT increased monotonically over the first four subperiods we consider; the odds of a VPT are 5.44 times higher in the 2000–2006 subperiod than in 1978–1982. The odds ratio then decreased, to 3.87 in 2007–2008. For the MCH-FP Area, the odds of a VPT increased between 1983–1992 and 1993–2000, but the increase was not as great as in the Comparison Area. There were comparable increases, in relative terms, in both areas between 1993–2000 and 2001–2006. However, between 2001–2006 and 2007–2008, the odds of VPT decreased in the Comparison Area but increased in the MCH-FP Area. In both areas, we see no evidence of a marked change in the likelihood of VPT during the subperiods when the Mexico City Policy was in effect (1983–1992 and 2001–2008) compared with when it was not.
Discussion
In our multivariate analyses, we find that variables that are associated with the desire to not to have another birth (soon)—very young or older maternal age, a greater number of living children, a recent multiple birth, and a preceding short IPI following a LB or a VPT—are associated with a greater likelihood of VPT. By contrast, variables that are associated with a desire to have a LB—for example, not having any living sons, the recent death of a child (particularly of a son), and a recent pregnancy that ended with a miscarriage or a stillbirth—are associated with a lower likelihood of VPT. Even if there is some underreporting of VPTs in our data, the ones that are reported are related to the explanatory variables in the ways we expect.
The finding of a high likelihood of VPT for very young women might seem surprising because it is believed that there is considerable pressure for young women to have a child immediately after marriage in Bangladesh (NIPORT, Mitra and Associates, and MEASURE DHS ICF International 2013). (Nonmarital childbearing is very rare in Bangladesh, so this would not account for the effect of young maternal age, as it does in some countries.) However, BDHS data show that in 2007, more than one in four women (27.4 %) with no children expressed a desire to have a child later (NIPORT, Mitra and Associates, and ORC Macro 2009); the percentage was nearly as high (23.3 %) about two decades earlier (Mitra et al. 1994). Such women are young, and they tend to use oral contraceptives or condoms, which have high use-failure. Method-specific likelihood of use-failure is higher for young or nulligravid women than for other women (Bairagi and Rahman 1996). If such women are highly motivated to delay their first pregnancy, they are likely to opt for VPT if they become pregnant before they want to. Furthermore, the intensity of the desire to delay the first birth appears to have increased over time. In 1993–1994, 1 in 11 women used contraception to delay the first birth; by 2007, the ratio was 1 in 5 (Mitra et al. 1994; NIPORT, Mitra and Associates, and ORC Macro 2009).
We find that women appear to have preferences about the spacing of pregnancies. After a LB, women prefer not to have another pregnancy within 15 months; and the shorter the interval, the more likely they are to terminate such pregnancies. The likelihood of VPT also increases as intervals become longer than 50 months, suggesting that many of these pregnancies were to women who did not want to have another child but had an unintended conception, perhaps because they were not using (effective) contraception or thought that they could no longer conceive.
Pregnancies after short intervals that began with a VPT have a very high likelihood of being terminated. The woman didn’t want to have a child the previous time she was pregnant and presumably doesn’t now, either; also, she has already demonstrated her willingness to have a VPT.
Excepting short IPIs that began with a miscarriage or a stillbirth, the likelihood of VPT is lowest for IPIs of 15–50 months long in the 1978–2000 time period and for IPIs of 27–74 months in the 2001–2008 period, suggesting that these are the IPI durations that women most preferred in each time period and that the preferred IPI duration increased between 1978–2000 and 2001–2008.
A number of our results are consistent with son preference, including the low likelihood of VPT if the woman didn’t have any living sons (a relationship also found by Bairagi (2001)) or if she recently experienced the death of a son, and the greater likelihood of termination after a short IPI that began with the birth of a son (compared with one that began with the birth of a daughter).
The higher the woman’s level of education, the more likely she is to terminate a pregnancy. Educated women in Bangladesh may be more likely to have unintended pregnancies because they are less likely than those with less education to use permanent methods of contraception and are more likely to use methods with higher failure rates, such as condoms, oral contraceptives, or traditional methods. In addition, educated women desire fewer children, and they are likely to perceive higher “costs” of having an unintended birth. Furthermore, educated women may have greater decision-making power and better access to information (e.g., regarding MR). We also find positive relationships, although not as strong, with the husband’s education and with the highest category of household space in the 1978–2000 period.
Muslims are less likely to terminate their pregnancies than are non-Muslims, presumably because the former are more conservative and attach more stigma to VPT.
In the 1978–1982 subperiod, pregnancies were much more likely to be terminated in the MCH-FP Area than in the Comparison Area. MR was provided by the MCH-FP Project during this time. Between 2002 and 2008, pregnancy termination was significantly less likely in the MCH-FP Area, consistent with that area having a more effective contraceptive method mix (and, therefore, fewer unintended pregnancies) than the Comparison Area (icddr,b 2011). However, in recent years, the VPT difference between the areas has narrowed, reflecting the relatively greater improvements in contraceptive use in the Comparison Area beginning in 2000, and possibly also the stagnation of the contraceptive use rate in the MCH-FP Area (while the intensity of desire to regulate fertility continued to increase). The recent increase in VPT in the MCH-FP Area may also reflect the fact that pregnancy-detection kits (which lead to an earlier awareness of being pregnant) have been increasingly used in that area in recent years.11
The increases in the likelihood of VPT seen over much of the study period likely reflect decreases in desired family size. Children have become more expensive in Bangladesh, and hence the costs of having an unintended birth are greater, leading to an increased intensity of desire to space and limit pregnancies. Indeed, many of the explanatory variables affecting the likelihood of VPT—particularly extremes of maternal age, number of living children, short IPIs that began with LBs, and the previous outcome being multiple births—have stronger effects in more recent years (2001–2008) than in earlier years. Furthermore, even when the explanatory variables are held constant, we generally see a trend of increasing likelihood of VPT over time in both areas. These increases may also be due to the increasing availability of MR, since this legal and relatively safe method reduces the total costs of VPT. We see no evidence of a marked difference in the likelihood of VPT during the times when the Mexico City Policy was in effect (1983–1992 and 2001–2008) compared with when it was not.
The results indicating an increased intensity of desire for pregnancy spacing and limitation are consistent with the fact that Bangladesh is rapidly transforming into a low-fertility country. Although it appears to be declining, unmet need for contraception is still fairly high in Bangladesh: 11.7 % in 2011 (NIPORT, Mitra and Associates, and MEASURE DHS ICF International 2013) (compared with 17 % in 2007 (NIPORT, Mitra and Associates, and ORC Macro 2009)). The desire of Bangladeshi couples to limit their family size may be even stronger in the future, with continued rapid social transformation and increased population crowding. Unmet need could continue to lead to more VPTs as pregnancies that might be prevented by use of effective contraception are instead terminated.
Analysis of data from a recent Matlab survey found that women who had VPTs were more likely to have used contraception prior to conception (35 %) than women who did not terminate their pregnancies (10 %) (Razzaque et al. 2011), showing that the former group was more likely to desire to prevent a conception. Ninety percent of the contraceptive users who had VPTs were using oral contraceptives or condoms; and for 93 % of these contraceptive users, conception occurred while they were using a method, implying that many VPTs are occurring because of contraceptive use-failure. Among those who had a VPT but were not practicing contraception before the conception, nearly one-half cited side effects or the inability to obtain a suitable method as the reason, and another 17 % said that they were amenorrheic. A smaller 1995 survey in Matlab showed similar findings (Bhuiya et al. 2001): 36 % of women who had VPTs thought that the pregnancy was due to failure of FP methods they were using.
Although contraceptive use continues to increase in Bangladesh, couples rely on less-effective spacing methods of contraception for fertility-limiting purposes, and appear to use VPT as a backup strategy to prevent exceeding their desired family size. Nationally in 2011, 41.9 % of currently married women aged 15–49 used oral contraceptives, condoms, or traditional methods, all of which have high rates of discontinuation or failure; slightly more than 11.2 % used injectables, which have low failure rates but, like oral contraceptives, have high discontinuation rates because of side effects; and only 6.9 % used female or male sterilization or IUDs, which have no or few contraception failures (NIPORT, Mitra and Associates, and MEASURE DHS ICF International 2013). Women over age 30 and those with two or more children, who are likely to be limiters, continue to use short-acting methods (NIPORT, Mitra and Associates, and MEASURE DHS ICF International 2013). To reduce the incidence of unintended pregnancies and hence to reduce the incidence of VPT, the FP program should aim to improve the mix of permanent and temporary contraceptives (including methods for amenorrheic women), reduce use-failure rates of temporary methods and the side effects of otherwise-reliable injectables, and increase the use of effective modern methods among fertility limiters. The FP program should also try to reach women with characteristics identified here as being associated with an increased likelihood of VPT.
Acknowledgments
Support was provided by the WHO Research Program on Sexual and Reproductive Health and by the Office of Population and Reproductive Health, Bureau for Global Health, U.S. Agency for International Development under the terms of Cooperative Agreement No. GPO-A-00-05-00027-00 awarded to the Extending Service Delivery project (a partnership among Pathfinder International, IntraHealth International, Inc., Management Sciences for Health, and Meridian Group International, Inc., that ended in 2011). The views expressed are those of the authors and do not reflect the opinions of their organizations or the funders. The authors thank Maureen Norton for her helpful suggestions.
Notes
We use the term “VPT” here rather than “induced abortion” because some terminations in Bangladesh—namely, those done by menstrual regulation (explained in the next section of this article)—are not considered to be abortions in Bangladesh.
The median and average durations of pregnancies that end in VPT are shorter than those for miscarriages and, of course, LBs and stillbirths. See Appendix A.
Abortion is a sensitive topic in Bangladesh; many of the restrictions on MR, particularly its availability only before pregnancy is clinically confirmed—and, in fact, its name—are to reinforce the perception of MR as something other than abortion. Nonetheless, in this article, we sometimes use the term “abortion” to refer to VPT to be consistent with use in the literature. What we call “abortions” includes MRs.
The Comparison Area contraceptive-use rate might be underestimated because data collection procedures were less rigorous there than in the MCH-FP Area (icddr,b 2011).
We also estimated a multinomial logistic regression explaining whether pregnancies end in VPT, miscarriage, or stillbirth, all relative to their ending in an LB. The conclusions about influences on VPT are identical to those presented here.
We initially explored in bivariate analyses how the effects of age and of number of living children on VPT differed across a number of subperiods of time. For both of these explanatory variables, the largest differences were between subperiods before and after 2001. In the multivariate analysis, we interact all explanatory variables with a dichotomous indicator for 2001–2008; the statistical significance of the coefficients of the interaction indicates whether the effects of the variable differ significantly between the two time periods. We also show the statistical significance of the total effects in 2001–2008 of all explanatory variables.
The duration of the IPI is computed as the amount of time between the date of previous pregnancy outcome and the estimated date of conception, which we compute as the date of last menstrual period (DLMP) before the index outcome plus two weeks. DLMP was not recorded in the Comparison Area before 2001. Also, there is no information on DLMP for women who entered the DSS area after their LMP. For the 42 % of cases with unknown DLMP, we estimated DLMP by subtracting the average outcome-specific duration of gestation for cases with known DLMP (8, 11, 33, and 36 weeks for VPT, miscarriage, stillbirth, and LB, respectively) from the date of the index outcome.
In exploratory analyses, we considered each two-year subperiod in the 2000s, but effects did not differ significantly across the first three of these.
The multivariate analysis controls for some variables correlated with older maternal age (e.g., number of living children). When we do not control for the other covariates, the effects are larger for older maternal ages than for the youngest ones.
The regression also includes an explanatory variable for first pregnancies (as part of the IPI set of variables). All first pregnancies are cases in which there are no living children, so the effects of these two variables need to be considered together. The no-living-children effect by itself is for cases in which the pregnancy is not the woman’s first but her previous pregnancies did not result in the LB of a child who is still alive.
These were first used for the MINIMAT project, which operates among a subset of women in the MCH-FP Area. They were so popular that other women in the area requested them (personal communication with K. Streatfield, October 27, 2011).
Appendix A: Outcome-Specific Durations of Pregnancy
Table 3 shows mean and median durations of pregnancy in weeks (measured from the estimated date of conception, computed as the date of the last menstrual period (DLMP) + 2 weeks), by outcome, for pregnancies during the period 1978–2008 for which DLMP was reported (58 % of all cases considered here). We also show data for MR and other methods of VPT for the years that the VPT method can be distinguished in the DSS (1989–2008, excluding 2001).
